Power prediction method and system combining multi-meteorological large model and WRF model

Abstract

The present invention provides a power prediction method and system combining a multi-meteorological large model and a WRF model, and relates to the technical field of power prediction of renewable energy power generation in power systems. When a prediction task is received, basic data is prepared; a preset model is driven according to global historical data to generate predicted meteorological data; the predicted meteorological data is combined with the predicted preparation data to generate combined predicted data; the combined predicted data is calculated by the WRF / WRF‑Solar model to generate mesoscale meteorological data; the target extraction number is obtained from the mesoscale meteorological data by bilinear interpolation; the historical data of new energy stations is fitted by a machine learning algorithm to obtain a training model; the target extraction data is input into the training model to generate power prediction data. By integrating the multi-meteorological large model with the predicted data from different meteorological data sources, the accuracy and reliability of the new energy power prediction can be improved.

Description

Technical Field

[0001] The present application relates to the technical field of power prediction of renewable energy power generation in power systems, and in particular to a power prediction method and system combining a multi-meteorological large model and a WRF model. Background Art

[0002] As the proportion of renewable energy in the power system continues to increase, accurately predicting the power generation of renewable energy such as wind and solar energy is of great significance for grid dispatching, power trading, and ensuring power supply stability. Existing renewable energy power prediction methods often rely on a single meteorological data source or model, and the prediction accuracy is limited by factors such as data quality and model complexity.

[0003] Therefore, how to improve the accuracy and reliability of new energy power prediction has become a technical problem that needs to be solved urgently. Summary of the invention

[0004] In order to improve the accuracy and reliability of renewable energy power prediction, the present application provides a power prediction method and system combining a multi-meteorological large model and a WRF model.

[0005] In the first aspect, the present application provides a power prediction method combining a multi-meteorological large model and a WRF model using the following technical solutions:

[0006] A power prediction method combining multiple meteorological large models and WRF model, comprising:

[0007] When receiving a forecasting task, basic data preparation is performed, and the basic data includes: global historical data, forecasting preparation data, and new energy station historical data;

[0008] Driving a preset model according to the global historical data to generate forecast meteorological data;

[0009] combining the forecast meteorological data with the forecast preparation data to generate combined forecast data;

[0010] The combined forecast data is operated by a WRF / WRF-Solar model to generate mesoscale meteorological data corresponding to the area of ​​interest;

[0011] Extracting target extraction data from the mesoscale meteorological data by bilinear interpolation;

[0012] Fitting the nonlinear relationship between meteorological data and power in the historical data of the new energy station through a machine learning algorithm to obtain a training model;

[0013] The target extraction data is input into the training model to generate power prediction data for the new energy station.

[0014] Optionally, the step of preparing basic data includes:

[0015] Obtain a preset meteorological large model combination, and use the corresponding fields in the meteorological large model combination as download restriction conditions to download global historical data from a preset meteorological data service source;

[0016] Obtaining a WRF / WRF-Solar model and downloading forecast preparation data in a preset meteorological data port according to field requirements in the WRF / WRF-Solar model;

[0017] The historical data of the new energy station is obtained, wherein the historical data includes all seasonal characteristic values.

[0018] Optionally, the step of driving a preset model according to the global historical data to generate forecast meteorological data includes:

[0019] Get the time constraints corresponding to the preset model;

[0020] Inputting the global historical data into the preset model to obtain a model output result;

[0021] Generate forecasted weather data in the model output according to the time constraint.

[0022] Optionally, the step of combining the predicted meteorological data with the prediction preparation data to generate combined prediction data includes:

[0023] Determine the field correspondence between the forecasted meteorological data and the forecast preparation data;

[0024] Generate a data processing strategy according to the field correspondence, the data processing strategy including: data clipping, data assembly and data conversion;

[0025] A preset data processing tool is obtained and the forecast meteorological data and the forecast preparation data are combined according to the data processing strategy to generate combined forecast data.

[0026] Optionally, the step of operating the combined prediction data through the WRF / WRF-Solar model to generate mesoscale meteorological data corresponding to the area of ​​interest includes:

[0027] generating a time configuration strategy according to the time information in the combined prediction data;

[0028] Generate coordinates and grid configuration strategies based on the latitude and longitude coordinate positions in the combined prediction data;

[0029] Determining a terrain file selection strategy according to the mesh refinement level in the prediction task;

[0030] Generate rate configuration strategy based on the power forecast frequency of new energy stations;

[0031] The mesoscale meteorological data corresponding to the area of ​​interest is generated by combining the WRF / WRF-Solar model with the time configuration strategy, the coordinate and grid configuration strategy, the terrain file selection strategy and the result evaluation configuration strategy.

[0032] Optionally, the step of extracting target extracted data from the mesoscale meteorological data by bilinear interpolation includes:

[0033] Obtaining grid point information in the mesoscale meteorological data;

[0034] Determine the target latitude and longitude and the target altitude according to the prediction task;

[0035] Determine target interpolation data in the grid point information by combining the target longitude and latitude and the target altitude in a bilinear interpolation manner;

[0036] Target extraction data is determined in the target interpolation data according to the power characteristics.

[0037] Optionally, the step of fitting the nonlinear relationship between meteorological data and power in the historical data of the new energy station by a machine learning algorithm to obtain a training model includes:

[0038] Determine a target characteristic value according to field information in the historical data of the new energy station;

[0039] Determine a training set, a validation set, and a test set according to the time length information in the historical data of the new energy station;

[0040] Inputting the target feature value into a machine learning algorithm in combination with the training set, the validation set, and the test set to determine the optimal training parameters and rounds by cross-validation;

[0041] A training model is determined using the optimal training parameters and rounds.

[0042] In a second aspect, the present application provides a power prediction system combining a multi-meteorological large model and a WRF model, wherein the power prediction system combining a multi-meteorological large model and a WRF model comprises:

[0043] A data preparation module is used to prepare basic data when receiving a prediction task, and the basic data includes: global historical data, prediction preparation data and new energy station historical data;

[0044] A forecasting meteorological data module, used for driving a preset model according to the global historical data to generate forecasting meteorological data;

[0045] A combined forecast data module, used for combining the forecast meteorological data with the forecast preparation data to generate combined forecast data;

[0046] A mesoscale meteorological data module, for computing the combined forecast data through a WRF / WRF-Solar model to generate mesoscale meteorological data corresponding to the area of ​​interest;

[0047] A target extraction data module, used for extracting target extraction data from the mesoscale meteorological data by bilinear interpolation;

[0048] A training module, used for fitting the nonlinear relationship between meteorological data and power in the historical data of the new energy station through a machine learning algorithm to obtain a training model;

[0049] A power prediction data module is used to input the target extraction data into the training model to generate power prediction data for the new energy station.

[0050] In a third aspect, the present application provides a computer device, comprising: a memory and a processor, wherein the processor executes the method described above when running computer instructions stored in the memory.

[0051] In a fourth aspect, the present application provides a computer-readable storage medium, comprising instructions, which, when executed on a computer, enable the computer to execute the method described above.

[0052] In summary, the present application includes the following beneficial technical effects:

[0053] When receiving a forecast task, this application prepares basic data; generates forecast meteorological data based on the preset model driven by global historical data; combines the forecast meteorological data with the forecast preparation data to generate combined forecast data; calculates the combined forecast data through the WRF / WRF-Solar model to generate mesoscale meteorological data; obtains the target extraction number from the mesoscale meteorological data through bilinear interpolation; fits the historical data of new energy stations through machine learning algorithms to obtain a training model; and inputs the target extraction data into the training model to generate power forecast data. By integrating multiple large meteorological models and forecast data from different meteorological data sources, the accuracy and reliability of new energy power forecasts can be improved. BRIEF DESCRIPTION OF THE DRAWINGS

[0054] Figure 1 It is a schematic diagram of the computer device structure of the hardware operating environment involved in the embodiment of the present application.

[0055] Figure 2 It is a flow chart of the first embodiment of the power prediction method combining multiple meteorological large models and WRF models in the present application.

[0056] Figure 3 It is a flow chart of the first embodiment of the power prediction method combining multiple meteorological large models and WRF model in the present application, using Pangu and Fuxi as meteorological large models;

[0057] Figure 4 This is a schematic diagram corresponding to the method of using bilinear interpolation in the first embodiment of the power prediction method combining a multi-meteorological large model and a WRF model of the present application;

[0058] Figure 5 It is a structural block diagram of the first embodiment of the power prediction system of the present application combining multiple meteorological large models and WRF models. DETAILED DESCRIPTION

[0059] In order to make the purpose, technical solution and advantages of the present application more clearly understood, the present application is further described in detail below through the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present application and are not used to limit the present application.

[0060] Reference Figure 1 , Figure 1 A schematic diagram of the computer device structure of the hardware operating environment involved in the embodiment of the present application.

[0061] like Figure 1 As shown, the computer device may include: a processor 1001, such as a central processing unit (CPU), a communication bus 1002, a user interface 1003, a network interface 1004, and a memory 1005. Among them, the communication bus 1002 is used to realize the connection and communication between these components. The user interface 1003 may include a display screen (Display), an input unit such as a keyboard (Keyboard), and the optional user interface 1003 may also include a standard wired interface and a wireless interface. The network interface 1004 may optionally include a standard wired interface and a wireless interface (such as a wireless fidelity (Wireless-Fidelity, Wi-Fi) interface). The memory 1005 may be a high-speed random access memory (Random Access Memory, RAM), or a stable non-volatile memory (Non-Volatile Memory, NVM), such as a disk storage. The memory 1005 may also be a storage device independent of the aforementioned processor 1001.

[0062] Those skilled in the art will understand that Figure 1The structure shown in the figure does not constitute a limitation on the computer device, and may include more or less components than shown in the figure, or combine some components, or arrange the components differently.

[0063] like Figure 1 As shown, the memory 1005 as a storage medium may include an operating system, a network communication module, a user interface module, and a power prediction program combining a multi-meteorological large model and a WRF model.

[0064] exist Figure 1 In the computer device shown, the network interface 1004 is mainly used for data communication with the network server; the user interface 1003 is mainly used for data interaction with the user; the processor 1001 and the memory 1005 in the present application can be set in the computer device, and the computer device calls the power prediction program combining the multi-meteorological large model and the WRF model stored in the memory 1005 through the processor 1001, and executes the power prediction method combining the multi-meteorological large model and the WRF model provided in the embodiment of the present application.

[0065] The present application embodiment provides a power prediction method combining a multi-meteorological large model and a WRF model, referring to Figure 2 , Figure 2 This is a flow chart of the first embodiment of the power prediction method combining multiple meteorological large models and WRF models in this application.

[0066] In this embodiment, the power prediction method combining the multi-meteorological large model and the WRF model includes the following steps:

[0067] Step S10: When a prediction task is received, basic data preparation is performed, and the basic data includes: global historical data, prediction preparation data, and new energy station historical data.

[0068] It should be noted that the steps for basic data preparation include: obtaining a preset large meteorological model combination, and using the corresponding fields in the large meteorological model combination as download restrictions to download global historical data from a preset meteorological data service source; obtaining the WRF / WRF-Solar model and downloading forecast preparation data in the preset meteorological data port according to the field requirements in the WRF / WRF-Solar model; and obtaining historical data of new energy stations, which includes all seasonal characteristic values.

[0069] In the specific implementation, the steps to prepare historical global data include: downloading historical data from meteorological data service sources such as DWD Genmany, NOAA US, ECMWF, GFS, JMA Japan, etc. The specific fields are based on the specific requirements of the big model, and you can learn more about the fields required by the big model according to the official documents of the big model.

[0070] The steps to download the forecast data set include: downloading weather forecast data from the US Weather Bureau's Global Forecast System (GFS), the European Centre for Medium-Range Weather Forecasts (ECMWF) and other weather forecast data services. The time frequency of the data should be kept consistent with the time frequency of the large model output results.

[0071] The steps for collecting historical data of new energy sites include: for wind farms, it is necessary to collect historical data of wind speed and power (output) of each wind turbine, with a time frequency of 15 minutes and a time range of more than 1 year, so as to include various wind speed conditions throughout the year.

[0072] For photovoltaic sites, it is necessary to collect historical data on irradiance and power (output) with a frequency of 15 minutes and a time range of more than 1 year, so as to include various wind speed conditions throughout the year.

[0073] Step S20: driving a preset model according to global historical data to generate forecasted meteorological data.

[0074] It should be noted that it is understandable that the steps of driving the preset model based on global historical data to generate predicted meteorological data include: obtaining the time restriction conditions corresponding to the preset model; inputting the global historical data into the preset model to obtain the model output results; and generating predicted meteorological data in the model output results according to the time restriction conditions.

[0075] In the specific implementation, according to the requirements of the large meteorological model, the data of the corresponding time is downloaded. Some large meteorological models are one time, and some are two time; they are assembled into corresponding data formats, some are npy data format, and some are nc data format; according to the requirements, the corresponding tool library of Python is used for assembly and conversion.

[0076] Step S30: Combine the forecast meteorological data with the forecast preparation data to generate combined forecast data.

[0077] It should be noted that the step of combining the predicted meteorological data with the forecast preparation data to generate combined forecast data includes: determining the field correspondence between the predicted meteorological data and the forecast preparation data; generating a data processing strategy based on the field correspondence, the data processing strategy including: data clipping, data assembly and data conversion; obtaining a preset data processing tool and combining the predicted meteorological data and the forecast preparation data according to the data processing strategy to generate combined forecast data.

[0078] It can be understood that the correspondence between the forecast data downloaded from the meteorological source and the forecast data fields of the large meteorological model is found; the forecast data downloaded from the meteorological source and the forecast data of the large meteorological model are cropped, assembled and converted through python tools, NC file processing tools, and GRIB file processing tools; the forecast data after cropping, assembly and conversion are ensured to meet the input requirements of the WRF (Weather Research and Forecasting) model and its solar energy dedicated module WRF-Solar at the regional scale, so as to drive it to perform refined meteorological simulation.

[0079] Step S40: Calculate the combined prediction data through the WRF / WRF-Solar model to generate mesoscale meteorological data corresponding to the area of ​​interest.

[0080] In a specific implementation, the steps of operating the combined prediction data through the WRF / WRF-Solar model to generate mesoscale meteorological data corresponding to the area of ​​interest include: generating a time configuration strategy according to the time information in the combined prediction data; generating a coordinate and grid configuration strategy according to the latitude and longitude coordinate positions in the combined prediction data; determining a terrain file selection strategy according to the grid fineness in the prediction task; generating a result evaluation configuration strategy according to the power prediction frequency of the new energy site; and generating mesoscale meteorological data corresponding to the area of ​​interest by combining the time configuration strategy, the coordinate and grid configuration strategy, the terrain file selection strategy, and the result evaluation configuration strategy through the WRF / WRF-Solar model.

[0081] It should be noted that according to the instructions in the official documents and related papers, the grid is delineated, and then the parameters related to WPS and WRF are configured, and the combined prediction data is input into the WRF calculation. The calculation results of the corresponding time and space are obtained. The specific configuration steps are as follows:

[0082] Configure the time configuration in the namelist.wps file in the WPS link and the namelist.input file in the WRF link according to the time information of the combined forecast data;

[0083] Configure the coordinate configuration and grid division configuration in the namelist.wps file in the WPS link and the namelist.input file in the WRF link according to the latitude and longitude coordinates of the new energy station;

[0084] Configure the terrain file selection in the namelist.wps file in the WPS stage according to the mesh refinement required for calculation;

[0085] Configure the physical parameter scheme in the namelist.input file in the WRF link according to the latitude and longitude information and terrain information of the new energy station.

[0086] Configure the result prediction frequency in the namelist.input file in the WRF link according to the power prediction frequency of the new energy station.

[0087] Run WPS first, and then use the results of WPS to drive WRF to get the mesoscale spatial meteorological forecast data under the corresponding grid.

[0088] Step S50: extracting from the mesoscale meteorological data by bilinear interpolation to obtain target extraction data.

[0089] It should be noted that the steps of extracting target extraction data from mesoscale meteorological data by bilinear interpolation include: obtaining grid point information in the mesoscale meteorological data; determining the target longitude and latitude and target altitude according to the prediction task; determining the target interpolation data in the grid point information by combining the target longitude and latitude and target altitude by bilinear interpolation; and determining the target extraction data in the target interpolation data according to the power characteristics.

[0090] Step S60: Fit the nonlinear relationship between meteorological data and power in the historical data of the new energy station through a machine learning algorithm to obtain a training model.

[0091] In a specific implementation, the nonlinear relationship between meteorological data and power in the historical data of new energy stations is fitted by a machine learning algorithm to obtain a training model, including: determining a target feature value according to field information in the historical data of the new energy station; determining a training set, a validation set, and a test set according to time length information in the historical data of the new energy station; inputting the target feature value into the machine learning algorithm in combination with the training set, the validation set, and the test set to determine the optimal training parameters and rounds by cross-validation; and determining a training model using the optimal training parameters and rounds.

[0092] It is understandable that the LightGBM framework commonly used in the GBDT algorithm is used for model training. Use at least one year of historical data for training to build a nonlinear relationship model between wind speed / radiance and power. The specific training steps are as follows: set different feature values ​​according to the fields of historical data; set the training set, validation set, and test set in a ratio of 7:2:1 according to the time length of historical data; obtain the best training parameters and rounds through cross-validation; use the best training parameters and rounds to train the data model.

[0093] Step S70: Input the target extraction data into the training model to generate power prediction data for the new energy station.

[0094] It should be noted that the integration of multi-meteorological large models and forecast data from different meteorological data sources in this embodiment aims to optimize the power generation prediction process of renewable energy sources such as wind and solar energy. It integrates multiple core links such as data preprocessing, large-scale model prediction, data fusion, key power parameter extraction and power curve fitting to ensure the high accuracy and practicality of the prediction results. First, the multi-meteorological large model is creatively deeply integrated with meteorological forecast data as an advanced data source to drive the WRF (Weather Research and Forecasting Model) and WRF-Solar model, which enhances the model's simulation ability for complex meteorological conditions; secondly, by introducing linear interpolation, key parameters such as wind speed and solar radiation intensity are accurately extracted from meteorological data in a vast geographic space. This strategy significantly improves the accuracy and efficiency of data extraction. The prediction method proposed in this embodiment not only greatly improves the accuracy and reliability of new energy power generation prediction, but also provides a solid prediction basis for the intelligent dispatching and operation management of the power system, and effectively promotes the optimization and upgrading of the energy structure and the sustainable development of the power industry.

[0095] In the specific implementation, taking a wind farm in Hunan as an example, the ERA5 global historical data set, GFS as the global forecast data set, Pangu and Fuxi as the meteorological large models are selected. The specific flow chart is attached. Figure 3 .

[0096] 1. Data preparation

[0097] Download ERA5 dataset

[0098] According to the requirements of Pangu and Fuxi meteorological models, download the ERA5 reanalysis dataset released by the European Centre for Medium-Range Weather Forecasts (ECMWF), and the download area is global. See Appendix 1 for specific field requirements:

[0099]

[0100] Among them, the pressure level requirements of the download fields in Table 1 are 1000, 925, 850, 700, 600, 500, 400, 300, 250, 200, 150, 100, and 50hPa.

[0101]

[0102] According to the requirements of the above table, use the python cds-api library to download the ERA5 hourly data on singlelevels and ERA5 hourly data on pressure levels datasets, the time is two moments before the forecast time. For example, if you need to forecast from 00:00 on May 1 to 00:00 on May 2, download the data at 12:00 and 18:00 on April 30.

[0103] Download GFS dataset

[0104] Download the forecast data set of the Global Forecast System (GFS) of the U.S. Weather Bureau. The specific data set is GFS0.25 Degree Hourly, which has a grid accuracy of 0.25 degrees (latitude and longitude) resolution and a time accuracy of 1 hour. Download the GFS forecast data for the next 24 hours at 00:00 on May 1, with one file for each hour.

[0105] Collect the wind speed and power data of the wind farm for more than one year (for example, from January 1 to March 30 of last year).

[0106] 2. Large meteorological model prediction

[0107] For the Pangu model, the downloaded ERA5 dataset (NetCDF format) needs to be processed as follows:

[0108] Use the netCDF4 library to read all variables in the ERA5 dataset.

[0109] A surface dataset with 4*721*1440 grids is constructed and converted into the npy data format, where 4 represents the four variables of the surface layer. For details, please refer to Table 2, and 721*1440 represents the number of grids obtained by dividing the global longitude and latitude by 0.25 degrees.

[0110] A high-altitude dataset with a grid number of 5∗13∗721∗1440 is constructed and converted into the npy data format, where 5 represents the five variables of the high-altitude layer, as shown in Table 1, 13 represents the 13 pressure layers, and 721∗1440 also represents the global grid number.

[0111] For the Fuxi large model, the downloaded ERA5 dataset (NetCDF format) needs to be processed as follows:

[0112] Use the xarray library to read all the variables in the ERA5 dataset.

[0113] Use ERA5 data from two time points with an interval of 6 hours.

[0114] The upper air layer data and surface data of the two moments are stitched together and saved in NetCDF data format.

[0115] Then use the data obtained by the above method as the input of the big model, use the short-term model of the big model (Pangu's pangu_weather_1 model and Fuxi's short model) for inference, and get the forecast data for the next 24 hours, corresponding to 24 forecast data files. The result obtained by Pangu is a file in npy data format. The result obtained by Fuxi is a file in NetCDF data format.

[0116] 3. Combined forecast data

[0117] Align the 24 Pangu and Fuxi prediction data files and the 24 GFS prediction files (grib) downloaded by time and merge them together. To merge files at one time, follow the steps below:

[0118] The prediction data of GFS is read through two pythons, xarray+cfgrib, so that it can be split according to different data layers (there are 35 data layers by default) and saved as NC format files.

[0119] Then read the data of some layers through NetCDF. According to the corresponding relationship in Table 3, only four layers of data need to be read, and the corresponding fields in the data layer are replaced with the prediction results of the large meteorological model and saved as NC format files.

[0120] Use the cdo tool to convert the 35-layer NC file into GRIB format data, and then rename the corresponding columns.

[0121] Use the cat command to concatenate the 35 layers of GRIB files into one GRIB file.

[0122] Through the above steps, the data at one moment can be spliced. By repeating 24 times, all 24 files can be spliced, and then 24 GRIB assembled prediction data can be obtained as the WRF initial field data.

[0123]

[0124] Extract the data fields in the netCDF file of the forecast results of the large meteorological model (as shown in the table above) and replace them with the corresponding fields in the corresponding data layer (typeOfLevel) in the Grib file of the GFS forecast data, paying attention to the restrictions of stepType and level.

[0125] WRF / WRF-Solar model calculation

[0126] 1) Configure WPS (WRF Pre-processing System)

[0127] Use the WRFDomainWizard tool to draw two layers of grids with the wind farm as the center (the grids here can be modified according to actual conditions, this is just an example) to configure the WPS parameters. See the table below for specific parameters.

[0128]

[0129] Configure two layers of grids as shown in the table above. The outer grid uses 30*30 grids, each grid size is 30km*30km, the inner grid uses 40*40 grid data, each grid size is 1 / 3 of the outer grid, that is, 10km*10km, and the relative position of the inner grid is (9,9) using the mercator coordinate system. The topographic map uses two precisions of '10m' (10 minutes is about 18km) and '5m' (5 minutes is about 9km). Use the default values ​​for other parameters, and configure the time parameters according to the result time of the third step. Modify the configuration file namelist.wps in WPS according to the above requirements.

[0130] Run WPS

[0131] Run geogrid.exe, link_grib.csh to index the prediction data, select Vtable.GFS as the Vtable for the operation, run ungrid.exe, run metgrid.exe, and finally obtain the intermediate operation results as the input of WRF.

[0132] Configure WRF (Weather Research and Forecasting)

[0133] The grid parameters and time parameters in WRF configuration are consistent with those in WPS, and the physical parameter configuration is as follows. This set of physical parameters is for reference only, and different physical parameters need to be set according to different terrain and time conditions.

[0134]

[0135] For other output formats, output every 15 minutes and merge the 96 files into one file. Modify the configuration file namelist.input in WRF according to the above requirements.

[0136] Run WRF

[0137] First, index the WPS run results to the WRF run directory. Then run real.exe, and then run wrf.exe. Here, wrf.exe can be run in parallel through mpirun. Finally, after running the output, the NC file of wrfout of the two-layer grid is obtained, and the output result of the inner grid is taken.

[0138] Extracting data

[0139] Use the two Python scripts netCDF4+wrf-python to read the data calculated by WRF, and obtain the wind speed / direction index of the corresponding wind turbine according to the latitude and longitude of the target wind farm wind turbine. The specific steps are as follows:

[0140] Read the five indicator data of u, v, ph, phb, and hgt in the NC file, among which ph, phb, and hgt are used to calculate the height, and u and v are used to calculate the wind speed and direction.

[0141] The specific altitude = (ph+phb) / 9.81+hgt. The altitude tensor obtained contains three dimensions: height, longitude, and latitude.

[0142] Read the two tensors u and v, which also contain three dimensions: altitude, longitude, and latitude.

[0143] Use the bilinear interpolation method, see the specific schematic diagram Figure 4 , calculate the wind speed and direction corresponding to the longitude, latitude and altitude. The specific calculation method of bilinear interpolation is as follows: first, perform two single linear interpolations in the x direction to obtain two temporary points R1(x, y1) and R2(x, y2), and then perform one single linear interpolation in the y direction to obtain P(x, y).

[0144]

[0145]

[0146]

[0147] 5) Save the difference results into csv or other database.

[0148] Model Training

[0149] Use the historical wind speed and power data in data preparation to train based on the LightGBM framework. The specific steps are as follows:

[0150] The data is divided into training set, validation set, and test set in a ratio of 7:2:1.

[0151] Set the training parameters of LightGBM, mainly setting the learning rate (0.02), loss function (mse, mean square error), and maximum training rounds (5000 times). The parameters here can be modified according to actual conditions.

[0152] Run the LightGBM code for training using the training set and validation set to get the best round.

[0153] Use the best round to train again to get the model file.

[0154] Use the test set to test the effect.

[0155] The above process can be carried out for multiple rounds, and then the training set, validation set, and test set can be used alternately and verified back and forth until the best effect is found.

[0156] Power Fitting

[0157] The model trained with LightGBM is input with wind speed data obtained from bilinear interpolation and output with final power data. Then these steps are performed on all wind turbines in the wind farm, and the power of all wind turbines is summed up at the time to obtain the power forecast data of the wind farm during this period.

[0158] When receiving a forecast task, this embodiment prepares basic data; generates forecast meteorological data based on global historical data to drive the preset model; combines the forecast meteorological data with the forecast preparation data to generate combined forecast data; calculates the combined forecast data through the WRF / WRF-Solar model to generate mesoscale meteorological data; obtains target extraction data from the mesoscale meteorological data through bilinear interpolation; fits the historical data of new energy stations through a machine learning algorithm to obtain a training model; and inputs the target extraction data into the training model to generate power forecast data. By integrating multiple large meteorological models and forecast data from different meteorological data sources, the accuracy and reliability of new energy power forecasts can be improved.

[0159] In addition, an embodiment of the present application also proposes a computer-readable storage medium, on which is stored a program for power prediction that combines multiple meteorological large models and a WRF model. When the program for power prediction that combines multiple meteorological large models and a WRF model is executed by a processor, the steps of the method for power prediction that combines multiple meteorological large models and a WRF model as described above are implemented.

[0160] Reference Figure 5 , Figure 5 This is a structural block diagram of the first embodiment of the power prediction system of the present application that combines multiple meteorological large models and WRF models.

[0161] like Figure 5As shown, the power prediction system combining multiple meteorological large models and WRF models proposed in the embodiment of the present application includes:

[0162] The data preparation module 10 is used to prepare basic data when receiving a prediction task, and the basic data includes: global historical data, prediction preparation data and new energy station historical data;

[0163] The forecast weather data module 20 is used to drive the preset model according to the global historical data to generate forecast weather data;

[0164] A combined forecast data module 30, for combining the forecast meteorological data with the forecast preparation data to generate combined forecast data;

[0165] A mesoscale meteorological data module 40 is used to operate the combined prediction data through the WRF / WRF-Solar model to generate mesoscale meteorological data corresponding to the area of ​​interest;

[0166] A target extraction data module 50 is used to extract target extraction data from the mesoscale meteorological data by bilinear interpolation;

[0167] A training module 60 is used to fit the nonlinear relationship between meteorological data and power in the historical data of the new energy station through a machine learning algorithm to obtain a training model;

[0168] The power prediction data module 70 is used to input the target extraction data into the training model to generate power prediction data for the new energy station.

[0169] It should be understood that the above is only an example and does not constitute any limitation on the technical solution of the present application. In specific applications, technicians in this field can make settings as needed, and the present application does not impose any limitation on this.

[0170] When receiving a forecast task, this embodiment prepares basic data; generates forecast meteorological data based on global historical data to drive the preset model; combines the forecast meteorological data with the forecast preparation data to generate combined forecast data; calculates the combined forecast data through the WRF / WRF-Solar model to generate mesoscale meteorological data; obtains target extraction data from the mesoscale meteorological data through bilinear interpolation; fits the historical data of new energy stations through a machine learning algorithm to obtain a training model; and inputs the target extraction data into the training model to generate power forecast data. By integrating multiple large meteorological models and forecast data from different meteorological data sources, the accuracy and reliability of new energy power forecasts can be improved.

[0171] It should be noted that the workflow described above is merely illustrative and does not limit the scope of protection of the present application. In practical applications, technicians in this field can select part or all of it according to actual needs to achieve the purpose of the present embodiment, and no limitation is made here.

[0172] In addition, for technical details not described in detail in this embodiment, please refer to the power prediction method combining multiple meteorological large models and WRF models provided in any embodiment of the present application, which will not be repeated here.

[0173] In addition, it should be noted that, in this article, the terms "include", "comprises" or any other variations thereof are intended to cover non-exclusive inclusion, so that a process, method, article or system including a series of elements includes not only those elements, but also includes other elements not explicitly listed, or also includes elements inherent to such process, method, article or system. In the absence of further restrictions, an element defined by the sentence "comprises a ..." does not exclude the existence of other identical elements in the process, method, article or system including the element.

[0174] The serial numbers of the above-mentioned embodiments of the present application are for description only and do not represent the advantages or disadvantages of the embodiments.

[0175] Through the description of the above implementation methods, those skilled in the art can clearly understand that the above-mentioned embodiment methods can be implemented by means of software plus a necessary general hardware platform, and of course by hardware, but in many cases the former is a better implementation method. Based on such an understanding, the technical solution of the present application, or the part that contributes to the prior art, can be embodied in the form of a software product, which is stored in a storage medium (such as a read-only memory (ROM) / RAM, a magnetic disk, or an optical disk), and includes a number of instructions for a terminal device (which can be a mobile phone, a computer, a server, or a network device, etc.) to execute the methods of each embodiment of the present application.

[0176] The above are only preferred embodiments of the present application, and are not intended to limit the patent scope of the present application. Any equivalent structure or equivalent process transformation made using the contents of the present application specification and drawings, or directly or indirectly applied in other related technical fields, are also included in the patent protection scope of the present application.

Claims

1. A power prediction method combining multiple meteorological large models and WRF models, characterized in that: include: When receiving a forecasting task, basic data preparation is performed, and the basic data includes: global historical data, forecasting preparation data, and new energy station historical data; Driving a preset model according to the global historical data to generate forecast meteorological data; combining the forecast meteorological data with the forecast preparation data to generate combined forecast data; The combined forecast data is operated by a WRF model to generate mesoscale meteorological data corresponding to the area of ​​interest; Extracting target extraction data from the mesoscale meteorological data by bilinear interpolation; Fitting the nonlinear relationship between meteorological data and power in the historical data of the new energy station through a machine learning algorithm to obtain a training model; Inputting the target extraction data into the training model to generate power prediction data for the new energy station; The basic data preparation includes: downloading the ERA5 data set. According to the requirements of the Pangu and Fuxi meteorological models, download the ERA5 reanalysis data set released by the European Centre for Medium-Range Weather Forecasts. The download area is global. Use the Python cds-api library to download the ERA5 hourly data on single levels and ERA5 hourly data on pressurelevels data sets. The time is two moments one day before the forecast time. Download the GFS data set. Download the forecast data set of the Global Weather Forecast System of the U.S. Meteorological Administration. The specific data set is GFS 0.25 Degree Hourly. The grid accuracy is 0.25 degree resolution and the time accuracy is 1 hour. Collect the wind speed and power data of the wind farm for more than one year. The steps of driving the preset model according to the global historical data to generate the forecasted meteorological data include: for the Pangu model, the following data processing needs to be performed on the downloaded ERA5 dataset in NetCDF format: using the netCDF4 library to read all the variables in the ERA5 dataset; constructing a 4 721 1440 grid number surface data set and converted to npy data format, where 4 represents the four variables of the surface layer, 721 1440 represents the number of grids obtained by dividing the global longitude and latitude by 0.25 degrees; construct a 5 13 721 The high-altitude data set with 1440 grid numbers is converted into npy data format, where 5 represents five variables of the high-altitude layer, 13 represents 13 pressure layers, and 721 1440 also indicates the number of global grids. For the Fuxi big model, the downloaded ERA5 dataset needs to be processed as follows: use the xarray library to read all variables in the ERA5 dataset; use the data of two moments with an interval of 6 hours in ERA5; splice the upper layer data and surface data of the two moments together and save them in NetCDF data format; use the data obtained by the above method as the input of the big model, use Pangu's pangu_weather_1 model and Fuxi's short model for inference, and obtain the forecast data for the next 24 hours, corresponding to 24 forecast data files; The step of combining the forecast meteorological data with the forecast preparation data to generate the combined forecast data includes: aligning the obtained 24 Pangu and Fuxi forecast data files and the downloaded 24 GFS forecast files according to the time, and merging the files at one time according to the following steps: Use xarray+cfgrib to read the prediction data of GFS, split it into different data layers and save it as NC format files; Then read the four layers of data through NetCDF, replace the corresponding fields in the data layer with the prediction results of the large meteorological model, and save them as NC format files; Convert the 35-layer NC file into GRIB format data through the cdo tool; Then rename the corresponding columns; Use the cat command to concatenate the 35 layers of GRIB files into one GRIB file; Through the above steps, the data at one moment can be spliced. By repeating 24 times, all 24 files can be spliced, and then 24 GRIB assembled prediction data can be obtained as the WRF initial field data.

2. The power prediction method combining multiple meteorological large models and WRF models according to claim 1 is characterized in that: The steps of preparing basic data include: Obtain a preset meteorological large model combination, and use the corresponding fields in the meteorological large model combination as download restriction conditions to download global historical data from a preset meteorological data service source; Obtaining a WRF model and downloading forecast preparation data in a preset meteorological data port according to field requirements in the WRF model; The historical data of the new energy station is obtained, wherein the historical data includes all seasonal characteristic values.

3. The power prediction method combining multiple meteorological large models and WRF models according to claim 1 is characterized in that: The step of driving a preset model according to the global historical data to generate forecast meteorological data comprises: Get the time constraints corresponding to the preset model; Inputting the global historical data into the preset model to obtain a model output result; Generate forecasted weather data in the model output according to the time constraint.

4. The power prediction method combining multiple meteorological large models and WRF models according to claim 1 is characterized in that: The step of combining the forecast meteorological data with the forecast preparation data to generate combined forecast data comprises: Determine the field correspondence between the forecasted meteorological data and the forecast preparation data; Generate a data processing strategy according to the field correspondence, the data processing strategy including: data clipping, data assembly and data conversion; A preset data processing tool is obtained and the forecast meteorological data and the forecast preparation data are combined according to the data processing strategy to generate combined forecast data.

5. The power prediction method combining multiple meteorological large models and WRF models according to claim 1 is characterized in that: The step of operating the combined prediction data through the WRF model to generate mesoscale meteorological data corresponding to the area of ​​interest includes: generating a time configuration strategy according to the time information in the combined prediction data; Generate coordinates and grid configuration strategies based on the latitude and longitude coordinate positions in the combined prediction data; Determining a terrain file selection strategy according to the mesh refinement level in the prediction task; Generate rate configuration strategy based on the power forecast frequency of new energy stations; The mesoscale meteorological data corresponding to the area of ​​interest is generated by combining the WRF model with the time configuration strategy, the coordinate and grid configuration strategy, the terrain file selection strategy and the result evaluation configuration strategy.

6. The power prediction method combining multiple meteorological large models and WRF models according to claim 1 is characterized in that: The step of extracting target extracted data from the mesoscale meteorological data by bilinear interpolation comprises: Obtaining grid point information in the mesoscale meteorological data; Determine the target latitude and longitude and the target altitude according to the prediction task; Determine target interpolation data in the grid point information by combining the target longitude and latitude and the target altitude in a bilinear interpolation manner; Target extraction data is determined in the target interpolation data according to the power characteristics.

7. The power prediction method combining multiple meteorological large models and WRF models according to claim 1 is characterized in that: The step of fitting the nonlinear relationship between meteorological data and power in the historical data of the new energy station by a machine learning algorithm to obtain a training model includes: Determine a target characteristic value according to field information in the historical data of the new energy station; Determine a training set, a validation set, and a test set according to the time length information in the historical data of the new energy station; Inputting the target feature value into a machine learning algorithm in combination with the training set, the validation set, and the test set to determine the optimal training parameters and rounds by cross-validation; A training model is determined using the optimal training parameters and rounds.

8. A power prediction system combining multiple meteorological large models and WRF models, characterized in that: Executing the method according to claim 1, the power prediction system combining multiple meteorological large models and WRF models comprises: A data preparation module is used to prepare basic data when receiving a prediction task, and the basic data includes: global historical data, prediction preparation data and new energy station historical data; A forecasting meteorological data module, used for driving a preset model according to the global historical data to generate forecasting meteorological data; A combined forecast data module, used for combining the forecast meteorological data with the forecast preparation data to generate combined forecast data; A mesoscale meteorological data module, used for computing the combined forecast data through a WRF model to generate mesoscale meteorological data corresponding to the area of ​​interest; A target extraction data module, used for extracting target extraction data from the mesoscale meteorological data by bilinear interpolation; A training module, used for fitting the nonlinear relationship between meteorological data and power in the historical data of the new energy station through a machine learning algorithm to obtain a training model; A power prediction data module is used to input the target extraction data into the training model to generate power prediction data for the new energy station.

9. A computer device, characterized in that: The device comprises: a memory and a processor, and when the processor runs the computer instructions stored in the memory, the processor executes the method according to any one of claims 1 to 7.

10. A computer-readable storage medium, characterized in that: The method comprises instructions, which, when executed on a computer, cause the computer to execute the method according to any one of claims 1 to 7.

Images